Respective roles of direct GHG radiative forcing and induced Arctic sea ice loss on the Northern Hemisphere atmospheric circulation

Abstract

The large-scale and synoptic-scale Northern Hemisphere atmospheric circulation responses to projected late twenty-first century Arctic sea ice decline induced by increasing Greenhouse Gases (GHGs) concentrations are investigated using the CNRM-CM5 coupled model. An original protocol, based on a flux correction technique, allows isolating the respective roles of GHG direct radiative effect and induced Arctic sea ice loss under RCP8.5 scenario. In winter, the surface atmospheric response clearly exhibits opposing effects between GHGs increase and Arctic sea ice loss, leading to no significant pattern in the total response (particularly in the North Atlantic region). An analysis based on Eady growth rate shows that Arctic sea ice loss drives the weakening in the low-level meridional temperature gradient, causing a general decrease of the baroclinicity in the mid and high latitudes, whereas the direct impact of GHGs increase is more located in the mid-to-high troposphere. Changes in the flow waviness, evaluated from sinuosity and blocking frequency metrics, are found to be small relative to inter-annual variability.

1 Introduction

Arctic sea ice decline in recent decades has been reported in observational studies (Stroeve et al. 2012a). Though the declining trends can be partially explained by internal climate variability (Zhang 2015), modeling studies have confirmed that the downward trend in Arctic sea ice is mainly caused by increasing Greenhouse Gases (GHGs) concentrations into the atmosphere (Kay et al. 2011; Stroeve et al. 2012b). The IPCC-AR5 report concluded that Arctic sea ice will continue to decrease and is projected to disappear in the middle of the twenty-first century, yielding an ice-free region during boreal summer season (Stocker et al. 2014). However large uncertainties still remain in the timing of an ice-free region (Stroeve et al. 2012a).

The short sea ice observational record, mainly since the development of satellite measurements in 1979, does not allow a full evaluation of the impacts of Arctic sea ice decline on decadal time scales. Moreover, the observed Arctic sea ice variation includes both internal and external influences, and the quantification of the respective contributions cannot be strictly assessed from observations alone. For these reasons, numerical approaches represent an alternative. In a number of modeling studies (cited below), General Circulation Models (GCMs) are used to perform idealized experiments in order to separate the respective factors (internal and external) affecting the climate system. The climate community has recently conducted a number of modeling studies, in order to evaluate and understand the impacts of the Arctic sea ice loss on the climate system (Alexander et al. 2004; Deser et al. 2010, 2015; Screen et al. 2013; Orsolini et al. 2012; Rinke et al. 2013; Blackport and Kushner 2016). Some studies have shown that Arctic sea ice decline can significantly affect the large-scale atmospheric dynamics at mid-to-high latitudes of the Northern Hemisphere by altering the storm-tracks, the jet stream (position and strength) and the planetary waves (Deser et al. 2010; Screen et al. 2013; Peings and Magnusdottir 2014). Nevertheless, large uncertainties still remain due to different factors: firstly, a low signal-to-noise ratio makes difficult do detect robust responses over surface variables such as the temperature and precipitation, in particular in winter (Screen et al. 2013; Suo et al. 2016). Second the differences in the experimental protocols lead to a large variety of responses for the different models. For example, Scinocca et al. (2009) and Blackport and Kushner (2016) study the response to Arctic sea ice loss through an experimental protocol that modifies the surface albedo; whereas Deser et al. (2015) uses a flux correction technique to reduce Arctic sea ice. Besides, the processes linking low Arctic sea ice extent with hemispheric winter circulation anomalies are not fully understood, for example, the mechanisms behind the atmospheric response (Gerdes 2006; Deser et al. 2007; Francis et al. 2009). Despite these difficulties, a robust response to sea-ice change affecting the large-scale atmospheric circulation in winter seems to emerge in several studies (Deser et al. 2010, 2015; Screen et al. 2013; Peings and Magnusdottir 2014; Blackport and Kushner 2016). This response consists of a weakening of the mid-latitude westerlies, coherent with the negative phase of the Northern Annular Mode (NAM, Thompson and Wallace 2000), and more locally with the negative phase of the North Atlantic Oscillation (NAO, Hurrell et al. 2003). However, the atmospheric circulation response to Arctic sea ice loss can be essentially non-linear (Petoukhov and Semenov 2010; Semenov and Latif 2015). These studies also reported that although maximum sea ice loss occurs in summer and autumn seasons, the response in terms of surface heat fluxes, temperature and precipitation peaks in winter, warming and moistening the polar atmosphere. Some studies have focused on the response of blocking events and extreme weather (Francis and Vavrus 2012, 2015; Cohen et al. 2014; Francis and Skific 2015; Barnes 2013, 2014; Cattiaux et al. 2016) but no consensus regarding a modification on the frequency/severity of them has been found so far. In addition to the reported changes on large-scale dynamics, the Arctic sea ice decline is expected to alter the vertical and horizontal tropospheric temperature gradients, influencing baroclinic instability and cyclogenesis. The storm-tracks response to future Arctic sea ice changes thus needs to be addressed. Harvey et al. (2014, 2015) linked the changes of storm-tracks to the changes in the meridional temperature gradient in the Coupled Model Intercomparison Phase 3 (CMIP3) simulations. A large inter-model spread in the North Atlantic is found to be associated with changes in the lower-tropospheric equator-to-pole temperature difference. Zappa et al. (2013) studied North Atlantic storm-tracks in the CMIP5 climate models and found that the mean response in winter is characterized by a tripolar pattern, with an increasing number of storms over the British Isles and a decrease over Greenland and Southern North Atlantic.

So far, the atmospheric response has often been studied using Atmospheric General Circulation Model (AGCMs) in which the future Arctic sea ice conditions are prescribed to the atmospheric model (Deser et al. 2010; Screen et al. 2013; Peings and Magnusdottir 2014).

Recently, two different experimental protocols have been applied on different versions of the National Center for Atmospheric Research coupled models, leading to similar results in terms of the atmospheric circulation response to Arctic sea ice loss. In the first protocol, Deser et al. (2015) (hereinafter DE15) use a flux correction technique to reduce Arctic sea ice; in the second, Blackport and Kushner (2016) use a similar technique as Scinocca et al. (2009) that modify the surface albedo. Both studies show that the AGCM approach leads to an atmospheric response, which is confined in the mid and high latitudes of the Northern Hemisphere (north of 30°N). This is mainly due to the absence of ocean–atmosphere feedbacks that could contribute to spread out the effects of Arctic sea ice decline beyond mid-high latitudes of the Northern Hemisphere. Besides, stand-alone atmosphere model approaches do not allow investigating the ocean dynamics response to Arctic sea ice decline, in particular in the North Atlantic sub-polar gyre (SPG) and on the Atlantic Meridional Overturning Circulation (AMOC). An open question is the role of the Arctic sea ice on the recent AMOC slow-down reported in observational records (Rahmstorf et al. 2015). This AMOC decrease is projected to continue in numerical experiments within CMIP5 database (Cheng et al. 2013) during the twenty-first century, and could in turn alter the large-scale atmospheric circulation through changes in the poleward heat transport, not only at mid-latitudes but also in tropical areas.

Actually, there are only a few modeling studies addressing the implications of Arctic sea ice loss on the climate system based on coupled global climate models (Orsolini et al. 2012; Rinke et al. 2013; Guo et al. 2014). In the above-mentioned study, DE15 designed an idealized experimental framework in which the seasonal cycle of Arctic sea ice is controlled in order to understand the role of ocean–atmosphere feedbacks in the atmospheric response to projected twenty-first century Arctic sea ice loss. DE15 compared a set of experiments performed with different ocean model configurations (fully-coupled, slab and no interactive ocean) of the Community Climate System Model version 4 (CCSM4). The fully coupled configuration leads to a global atmospheric response, exhibiting an equatorial symmetry and equatorward shift of the intertropical convergence zone (ITCZ) related to changes in the Hadley circulation. Their results show that Arctic sea ice loss and GHG-induced warming have competing effects on the meridional temperature gradient in middle latitudes. This could explain the non-significant latitudinal shift of the Northern Hemisphere jet stream, contrarily to the Southern Hemisphere westerlies that significantly migrate poleward (Barnes and Polvani 2013).

The innovative protocol of DE15 highlighted the importance of ocean–atmosphere feedbacks in the climate system response to projected Arctic sea ice decline. However, the response described above can be model-dependent, and it is therefore helpful to extend such a protocol to other climate models. The purpose of this study is to test the robustness of the response found by DE15, through a similar protocol applied to the CNRM-CM5 climate model (Voldoire et al. 2013). In our approach, the seasonal cycle of Arctic sea ice is controlled by applying a flux correction to the ocean component only over the regions covered by sea ice. Unlike DE15, which used a flux correction of the longwave radiative flux, we compute the correction term from the total non-solar flux at the ocean surface (sum of longwave, latent and sensible heat fluxes) to add or remove heat from the ocean, in order to melt Arctic sea ice under present conditions (end of the twentieth century), or reform Arctic sea ice under future conditions (end of the twenty-first century). These idealized experiments allow us to separate the relative effects of Arctic sea ice loss and increased GHGs, as well as to test the possible additivity of both effects. Beyond the large-scale atmospheric circulation, we will focus on the synoptic-scale variability, including changes in the Northern Hemisphere baroclinicity, storm-tracks, blockings and flow sinuosity.

The paper is organized as follows. In Sect. 2 we describe the coupled model, as well as the experimental protocol and its validation. Section 3 presents the results regarding the atmospheric response to the respective role of increased GHGs and Arctic sea ice decline. Then we focus on changes in blocking occurrence, flow sinuosity and baroclinicity, showing the Eady Growth Rate parameter. The North Atlantic and North Pacific storm-tracks responses are analyzed at the end of Sect. 3. Summary and discussion are presented in Sect. 4.

2 Methods

2.1 The CNRM-CM5 coupled model

In this study we use the fully coupled model CNRM-CM5 developed by the CNRM-CERFACS modeling group (Voldoire et al. 2013). CNRM-CM5 includes the atmospheric model ARPEGE-Climat version 5.2 at a horizontal resolution of 1.4° and 31 vertical levels (T127 triangular truncation) (Déqué et al. 1994), derived from the ARPEGE/IFS operational weather prediction model maintained by Météo-France and the European Centre for Medium-Range Weather Forecasts (ECMWF). The surface component is the SURFEX module, which is embedded in ARPEGE and includes three surface schemes for natural land, inland water (lakes) and sea/ocean areas based on the ISBA model (Noilhan and Planton 1989). The ocean component of CNRM-CM5 is the NEMO model version 3.2 (Madec 2008). The oceanic resolution is 1° on an ORCA1 triangular grid with 42 vertical levels (18 in the upper 250 m). The sea ice model GELATO version 5.7 (Salas y Mélia 2002) is embedded in NEMO. The system ARPEGE-SURFEX is coupled to NEMO-GELATO through the OASIS coupled version 3 (Valcke 2013). More details of the model components and performance can be found in Voldoire et al. (2013).

2.2 Experimental protocol

In order to understand and separate the relative role of Arctic sea ice loss and global warming induced by GHGs increasing, we performed four idealized experiments summarized in Table 1. We first conduct two control experiments named CTL20 and CTL21 for the present and future climate, in which the radiative forcing as well as the aerosols are constant throughout the integration and set to 1985 and 2085 values, respectively. We then carry out two additional sensitivity experiments in which a flux correction is applied to artificially either melt (ICE21) or keep (ICE20) the Arctic sea ice under either 1985 or 2085 radiative forcings, respectively. In other words, ICE21 (ICE20) consists of an experiment in which the radiative forcing corresponds to 1985 (2085) with Arctic sea ice conditions of the 2070–2099 (1970–1999) period as estimated from the five CNRM-CM5 RCP8.5 (historical–HIST) runs. ICE21 and CTL20 (ICE20 and CTL21) are initialized from one member from the HIST (RCP8.5) ensemble on the 1st January 1985 (2085). The four experiments are run for 200 years each and we discard the first 100 years in our analysis, as detailed in the rest of the paper.

Table 1 Summary of the four idealized experiments described in Sect. 2.2

In the ICE-type experiment the seasonal cycle of Arctic sea ice is controlled by applying a non-solar flux correction term to the ocean model. The reason to correct only the non-solar term is due to the treatment in the NEMO model, which only differentiates the solar from the non-solar flux component, the latter including longwave and turbulent (latent and sensible) fluxes.

Figure 1 shows the RCP8.5–HIST (thereafter ΔRCP, the period 2070–2099 is used for RCP8.5 and the period 1970–1999 is used for HIST) differences for Arctic sea ice concentration (Fig. 1a) and the non-solar flux at the surface (Fig. 1b), as well as the non-solar flux correction applied in ICE21 (Fig. 1c) and ICE20 (Fig. 1d) for the months of January, April, August and October, which are representative of the different seasons. The sea ice retreat is important (more than 60%) in winter and spring over the marginal seas (Fig. 1a): Hudson Bay, Barents and Kara Seas for the Atlantic side, and Bering, Chukchi and Sea of Okhotsk for the Pacific. In contrast, in summer and autumn the Arctic Ocean is nearly ice-free at the end of the twenty-first century. To estimate the non-solar flux correction used for ICE21 and ICE20 simulations, we first compute the ΔRCP differences of the non-solar flux for each month of the year. The non-solar flux response exhibits a dipole structure, with strong upward flux in regions of sea ice loss and downward flux in regions directly south of this upward flux (Fig. 1b). This dipole is coherent with other studies (Deser et al. 2010; Screen et al. 2013) and will be further discussed in Sect. 3.1. We do not apply directly the raw non-solar flux differences from Fig. 1b to perform ICE21 experiment, because when proceeding this way, sea ice conditions in the Arctic turn out to be largely overestimated compared to the target period (here RCP8.5). This can be explained by the numerous non-linear feedbacks processes associated with Arctic sea ice loss. We thus follow the methodology describe in DE15 and use the ΔRCP differences from Fig. 1b as a first guess for iteration to find the optimal multiplication factor β, such as the correction non-solar flux*β fits as best as possible the sea ice concentration and volume from the target period. The β–factor is estimated empirically from multiple simulations (see the supplementary material S1 for more details) and the value β = 0.6 is found to be the one that best approximate ICE21 to RCP8.5 sea ice conditions in 2070–2099. The resulting correction term (non-solar flux*β) is displayed in Fig. 1c. This correction term is positive (ocean gains heat to melt sea ice and compensate the net upward flux in response to sea ice loss), monthly-varying and is only applied on grid points where the decrease of Arctic sea ice is greater than 10%, according to Fig. 1a. Locally this correction can be very strong (more than 60 W m⁻² for most of the regions). Globally averaged (as indicated on Fig. 1c), it is in agreement with the total radiative forcing in RCP8.5 (for instance, for the month of January the correction is 1.3 W m⁻²).

Fig. 1

a Sea ice concentration difference between the RCP8.5 (2070–2099) and HIST (1970–1999) simulations. Here the ensembles mean is computed from 5 members. Units are in % of change. b Non-solar flux differences (in W m⁻²) between the RCP8.5 (2070–2099) and HIST (1970–1999) simulations. c, d Non-solar flux correction (in W m⁻²) applied respectively in ICE21 and in ICE20 experiments. Fluxes are defined positive in the downward direction. For clarity purpose, the label bar is set the same for the four figures. The values located on the top-left of c, d figures correspond to the globally averaged flux correction (in W m⁻²)

The same method is applied to estimate the flux correction for the ICE20 experiment, but in that case the correction term is negative (heat lost by the ocean to produce sea ice) (Fig. 1d). In ICE20, the Arctic sea ice target is the HIST simulation (1970–1999 period). Similarly to ICE21, the value of β that best matches HIST sea ice conditions is 0.7, indicating that the correction needed for ICE21 and ICE20 is not symmetric. Considering global spatial averages, in ICE21 (Fig. 1c) the flux correction for January, April, August and October are 1.30, 0.51, 0.11 and 0.66 W m⁻² respectively. These values are slightly weaker than those exhibited by the ICE20 correction (−1.56, −0.61, −0.13, −0.68 W m⁻², in Fig. 1d), indicating that more energy is needed to produce sea ice starting from 2085 than to melt sea ice starting from 1985.

For the control simulation CTL20, a small flux correction is also applied. Indeed the idea is to bring the Arctic sea ice conditions for the selected period in HIST (1970–1999) close to the one from the representative year (1985) that is used to (1) initialize the idealized simulations and (2) prescribe the GHG fixed concentration. In order to take into account the effects of the small increase of GHGs concentration during the period 1970–1999, a small correction term of the non-solar flux is thus necessary (not shown). It is estimated from the difference between the representative year 1985 and the temporal mean for the 1970–1999 period.

Figure 2 summarizes the experiments and how they allow separating the effect of Arctic sea ice loss from the effect of increasing GHGs:

Fig. 2

Schematic representation of the 4 idealized experiments performed in this work and how to use them to estimate the relative effects of GHGs increasing and Arctic sea ice decline. ICE21–CTL20 Arctic sea ice loss with present GHGs conditions (constant at year 1985). ICE20–CTL20 increased GHGs with present Arctic sea ice (1970–1999). CTL21–CTL20 effect of both increased GHGs and Arctic sea ice loss. CTL21–ICE20 Arctic sea ice loss with future GHGs conditions (constant to year 2085). CTL21–ICE21 increased GHGs with future Arctic sea ice (2070–2099)

• The difference between ICE21 and CTL20 evaluates the effect of Arctic sea ice loss under present GHG conditions.

• The difference between ICE20 and CTL20 estimates the effect of increased GHGs under present Arctic sea ice conditions.

• The difference between CTL21 and CTL20 should match the ΔRCP difference, in which both effects are taking into account.

In the following, the Arctic sea ice loss effect (ICE21–CTL20) is referred to as “ICE” and the GHG direct radiative effect (ICE20–CTL20) is noted “GHG”. The total effect (CTL21–CTL20) is named “GHG + ICE”. The difference between RCP8.5 and HIST simulations is noted “ΔRCP” and corresponds to the total response to climate change in the CNRM-CM5 model. It is important to recall that ΔRCP is not the same as “GHG + ICE”, since the former is the difference between two transitory simulations, as GHGs forcing is varying with time. Finally, the statistical significance of the response is assessed using a two-sided Student’s t test.

2.3 Evaluation of the experimental protocol

Figure 3 shows the annual cycle of Arctic sea ice extent (Fig. 3a) and volume (Fig. 3b) for HIST (1970–1999), RCP8.5 (2070–2099), ICE21 and CTL20 simulations. For the sake of comparison, the experiments ICE_coupled_21 and ICE_coupled_20 from DE15 have been added. The sea ice extent and volume in CTL20 are in good agreement with the HIST ensemble, indicating that the experimental protocol has worked satisfactorily. Similar conclusions can be made when comparing ICE21 and RCP8.5, except for a slight overestimation of ICE21 during spring season. The experiments CTL21 and ICE20 are also in agreement with their respective targets (RCP8.5 and HIST), though a slight overestimation of the sea ice volume is present in ICE20 throughout the year (not shown). Figure 3 shows that the experimental protocol implemented in this work leads to very similar results as in DE15 for the sea ice extent and volume. Figure 4 represents the sea ice concentration pattern for ICE21 and CTL20 experiments in March and September. For ICE21, the green line corresponds to the RCP8.5 sea ice edge (averaged over the period 2070–2099), while in CTL20 this line corresponds to the HIST one (averaged over the period 1970–1999). Both RCP8.5 and HIST sea ice concentration edges have been determined from the ensembles mean. In March, ICE21 shows a slight overestimation of sea ice concentration compared to RCP8.5, consistent with Fig. 3. In September, the Arctic is practically an ice-free region for both RCP8.5 and ICE21. In CTL20, sea ice concentration is in good agreement with the HIST ensemble mean, also consistent with Fig. 3. Finally, the bottom row of Fig. 4 shows that the sea ice concentration difference between ICE21 and CTL20 captures very well the ΔRCP (in green contour).

Fig. 3

a Annual cycle of the Arctic sea ice extent (in 10⁶ km²) for: HIST (1970–1999) in blue, RCP8.5 (2070–2099) in red, ICE21 in black, CTL20 in dashed black, ICE_coupled_21 in dashed green and ICE_coupled_20 in green (from DE15). The shading corresponds to the inter-member spread computed from 5 members for both HIST and RCP8.5. For ICE21 and CTL20 the last 100 years have been used as explained in the text. b Same for the Arctic sea ice volume (in 10³ km³)

Fig. 4

Sea ice concentration in ICE21 (top) and CTL20 (middle) and their differences (ICE21–CTL20, bottom) in March and September (in %). The green line correspond to the RCP8.5 ensemble mean (2070–2099) and the HIST ensemble mean (1970–1999) sea ice concentrations respectively in ICE21 and CTL20 plots

Note that in the first 10 years of each experiment, sea ice concentration and volume largely vary in response to the flux correction imposed at the surface ocean. Stabilization occurs after about 20 years (not shown). However, the timescales involved in the coupled model adjustment towards the equilibrium state are much longer, which is suggestive of active processes associated with ocean circulation. The stabilization of the experiments is evaluated through the Atlantic Meridional Overturning Circulation (AMOC), computed here as the maximum of the ocean meridional stream function at 30°N latitude. Figure 5 shows the AMOC time series for the period 1970–1999 in the HIST ensemble, for 2070–2099 in the RCP8.5 ensemble and for the 200 simulated years of ICE21 and CTL20. Comparing HIST and RCP8.5, there is a significant decrease of the AMOC, ranging from 15 to 9 Sv in response to global warming in the RCP8.5 scenario. This is consistent with AMOC slow-down reported in previous studies (Cheng et al. 2013; Srokosz et al. 2012). In ICE21, the AMOC decreases from ~16 to ~11 Sv in the first 60 years of simulation, and seems to reach the equilibrium afterwards. The explanation of this AMOC reduction in ICE21 is not straightforward. On the one hand, the experimental protocol itself can induce an AMOC decrease by the non-solar flux correction. On the other hand, the prompt sea ice melting can alter heat budgets and fresh water fluxes over the deep convection regions. Separating both effects is not an easy task, and more analysis or even additional simulations are needed to elucidate the reasons of the AMOC decrease. But this is beyond the scope of this study, which focuses on the atmospheric response. Note that CTL20 stands on the upper limit of the HIST ensemble due to the fact that: (1) the member of HIST used to initialize CTL20 exhibits a strong AMOC; (2) in CTL20 GHGs concentrations are set and maintained constant to 1985. These values are smaller than those averaged over 1970–1999 in HIST. Hence the CTL20 climate is slightly colder than the HIST counterpart, which could explain in part the stronger AMOC in CTL20; (3) the Southern ocean is characterized by long-term internal variability (Latif et al. 2013). Figure 5 shows that the AMOC equilibrium state is reached after about 100 years for both ICE and CTL simulations and justifies the discard of the first 100 years of the experiments in the rest of the paper.

Fig. 5

Time series for the Atlantic meridional overturning circulation (AMOC, units in Sv) at 30°N: for the HIST ensemble for the period 1970–1999 in blue, the RCP8.5 ensemble for the period 2070–2099 in red, the 200 years of ICE21 simulation (black solid line) and the 200 years of CTL20 simulation (dashed black line). The red and blue shadings indicate the 5-members spread computed as one standard deviation

We have performed additional analysis to validate the experimental protocol, such as the comparison between ΔRCP and “GHG + ICE” responses in terms of 2 m temperature and precipitation for winter (DJF) and summer (JJA) seasons (see Fig. S2 and S3 in the supplementary material). These 2 figures show the good agreement between the responses, except a slight overestimation of temperature and precipitation anomalies in the GHG + ICE effect. There are several hypotheses to explain this overestimation: (1) GHG + ICE are roughly stabilized climates, whereas in ΔRCP the radiative forcing varies with time. (2) Arctic sea ice loss includes several processes and feedbacks that lead to non-linear responses: in the RCP8.5 the sea ice is decreasing progressively, while our experiments correspond to an abrupt sea ice loss.

Table 2 summarizes some of the regional ICE responses obtained in this study and compares them to the values obtained in DE15. In general, there is a very good agreement with the values obtained in DE15, in despite of using different coupled models and slightly different experimental protocols. This gives robustness and confidence in the response to Arctic sea ice loss previously identified in DE15 and also obtained in this study. These responses will be more detailed in the next sections.

Table 2 Same table as in DE15 (see Table 2 in DE15) for the ICE effect in CNRM-CM5 and the ΔICE_coupled in DE15 for the season DJF

We show here that our experimental protocol is successful in reproducing the Arctic sea ice conditions at the end of twenty-first century for a radiative forcing corresponding to the end of the twentieth century. Despite the locally strong flux correction applied, the model behaves satisfactorily and an equilibrated state is obtained after 20 years for the atmospheric (surface) fields. In the ocean, the equilibrium state seems to be reached after 100 years of simulation. This protocol also shows that the simulated climate response ΔRCP can be reproduced to study the combined effects of GHGs direct radiative forcing and Arctic sea ice decline. In the next section, we investigate these relative roles on the Northern Hemisphere large-scale atmospheric dynamics and the associated synoptic variability.

3 Results

3.1 Surface heat flux response to Arctic sea ice loss

We first analyze the local response in terms of surface heat fluxes when the effect of Arctic sea ice decrease is isolated (ICE). Figure 6 shows the ICE response for the shortwave (Fig. 6a), longwave (Fig. 6b), latent (Fig. 6c) and sensible (Fig. 6d) heat fluxes for different seasons (DJF, MAM, JJA and SON) (Note that contrarily as in Fig. 1 fluxes are counted positively upward). A strong anomalous downward shortwave flux (more than −60 W m⁻²) is found in boreal summer (also in spring but to a lesser extent) due to the almost disappearance of the sea ice at the end of twenty-first century, leading to an enhanced solar radiation absorbed by the ocean surface (the high albedo of the ice is replaced by the relative lower albedo of the sea). This increased absorbed flux causes a warming of sea surface temperature (SSTs) (see Fig. S1 in the supplementary material), further contributing to sea ice melting during the summer. Except for the summer season, the total heat flux response is mainly controlled by the turbulent heat fluxes (latent and sensible), with a small contribution of the longwave component. Latent and sensible fluxes exhibit a maximum response (more than +50 W m⁻², Fig. 6c, d) over the marginal seas regions where the sea ice retreat is localized (Fig. 1a). A meridional dipole structure for the sensible (and to a lesser extent for the latent) heat flux is observed, with downward anomalies located southwards. As described in Deser et al. (2010) and Screen et al. (2013), this can be explained by the advection of warmer air from the marginal seas due to sea ice retreat, which is then transferred to the ocean.

Fig. 6

ICE response for the surface heat fluxes: a Shortwave heat flux, b longwave heat flux, c latent heat flux and d sensible heat flux. The response has been decomposed in four different seasons, considering DJF, MAM, JJA and SON. The ICE response is defined as the difference ICE21–CTL20 simulations, as explained in the text. Units are W m⁻². Dotted areas represent statistically significant differences according to a t test at the 5% confidence level

It is important to mention that even if Arctic sea ice loss is the largest in summer and autumn (Fig. 1a), the turbulent flux response is maximum in winter (Fig. 6c, d). This temporal delay between maximum sea ice decrease and surface response exists because the air-sea temperature contrast is the strongest in wintertime [similar results are shown in Deser et al. (2010, 2015) and Screen et al. (2013)], enhancing the turbulent local exchanges. Deser et al. (2010) shows that this delayed response is also observed in 2-meter temperature and precipitation, which are maximal in winter.

3.2 Large-scale atmospheric circulation response

3.2.1 Near-surface response

In this section, we investigate the low-level atmospheric circulation response with a focus on winter (DJF) and summer (JJA) seasons. The decomposition of the total effect GHG + ICE, into GHG and ICE effects as well as the ΔRCP, on sea level pressure (SLP) and zonal wind at 850 hPa (U850) are shown in Figs. 7 and 8 for winter and summer respectively. For both Figs. 7 and 8 the GHG + ICE and ΔRCP responses are very close, confirming once again the good functioning of the experimental protocol.

Fig. 7

a, e ΔRCP, b, f GHG + ICE, c, g GHG, d, h ICE effects (shading) for (top) sea level pressure (SLP, units in hPa) and (bottom) zonal wind at 850 hPa (U850, units in m s⁻¹) for winter (DJF) season. We recall that GHG + ICE effect is computed as CTL21–CTL20, the GHG effect as ICE20–CTL20 and the ICE effect is ICE21–CTL20. The green contours correspond to the climatological sea level pressure (contour interval is 5 hPa) and zonal wind at 850 hPa (contour interval is 10 m s⁻¹) computed from the CTL20 simulation. Dotted areas represent statistically significant differences according to a t test at the 5% confidence level

Fig. 8

Same as for Fig. 7 but for summer (JJA) season

The winter SLP GHG + ICE response exhibits a marked regional behavior with a significant decrease over the North Pacific (more than 5 hPa), North America and central Arctic, and a significant increase over Greenland, the North Atlantic, the Euro-Mediterranean region, central Asia and over the Himalayas (Fig. 7b). The separation in GHG (Fig. 7c) and ICE (Fig. 7d) effects suggests that Arctic sea ice loss contribution primarily drives the strong SLP decrease over the North Pacific, North America and central Arctic, as well as the increase over Greenland (Fig. 7d). As a consequence of the sea ice retreat, the SLP anomalies are positive over the Northern Eurasia in ICE. This is coherent with lots of studies (Mori et al. 2014; Vihma 2014; Cohen et al. 2014), which showed that the SLP increase over Eurasia in the past decades was mainly explained by anomalous sea ice in the Barents-Kara seas. Over the North Atlantic and Northern Eurasia, GHG and ICE have opposite impacts (Fig. 7c, d). In the North Atlantic, GHG induces positive SLP anomalies, whereas the ICE response projects onto the negative NAO pattern. Moreover GHG radiative forcing dominates the response in the North Atlantic subtropics and in the Mediterranean region. This results in a non-significant response in the North Atlantic when both effects are included in GHG + ICE (Fig. 7b). The same conclusions can be drawn for Northern Eurasia, which reflects a non-significant response in GHG + ICE, except over the Icelandic low. Our results suggest that in wintertime, the GHG effect dominates the total SLP response in the subtropics, while the ICE effect controls the response at higher latitudes.

The GHG + ICE response in DJF for U850 (Fig. 7f) shows significant values over the Pacific, with strong strengthening of the mid-latitude westerlies. In the North Atlantic no significant response is detected. U850 decreases over the Mediterranean region but is enhanced over Northern Europe. ICE induces a southward shift of the mid-latitude westerlies (Fig. 7h), which is consistent with the negative phase of the NAM and NAO patterns (see also Table 2). The response to GHG is less zonally homogeneous. It is characterized by an enhancement of westerly winds in the North Pacific; whereas in the North Atlantic a northward shift counterbalances the ICE effect (Fig. 7g). The respective SLP and U850 responses in wintertime highlight the different behavior between North Atlantic-European and North Pacific regions, as discussed in Dethloff et al. (2006). In the North Atlantic and European sector, ICE and GHG effects cancel each other, resulting in a non-significant response in GHG + ICE, while GHG effect dominates the total response over the Mediterranean basin. In the North Pacific, ICE and GHG responses have the same sign, which leads to strong SLP and U850 anomalies. The non-significant pattern found in the GHG + ICE effect in the North Atlantic is consistent with DE15, who showed the cancellation between GHG and ICE impacts in the CCSM4 model. The NAO-like pattern response in ICE is also in agreement with DE15, Screen et al. 2013 and Peings and Magnusdottir (2014). Similarly, Barnes and Polvani (2015) show that the North Atlantic jet stream exhibits no significant response to the Arctic Amplification in winter.

In summer (Fig. 8), the SLP GHG + ICE response displays positive values over North America, the North Atlantic and large part of Eurasia, and weak negative values over central Arctic (Fig. 8b). These changes appear to be clearly dominated by the GHG effect (Fig. 8c). The GHG + ICE U850 response is mainly characterized by a poleward shift of the westerlies in the North Atlantic, which is due to the GHG effect (Fig. 8f, g). Similar results have also been shown in Barnes and Polvani (2015). Non-significant U850 changes are observed in ICE, corroborating the fact that although maximum sea ice loss occurs in summer, its impact on air-sea fluxes and on atmospheric circulation are observed in winter.

3.3 Vertical structure of the response

We investigate now the vertical structure of the atmospheric response through the zonally averaged Northern Hemisphere air temperature (TA) and zonal wind component (UA) as a function of the pressure levels for the winter season (Fig. 9). The GHG + ICE response in winter for TA (Fig. 9b) exhibits a strong warming (more than 6 K) northern than 60°N confined between the surface and 600 hPa, which corresponds to the so-called Arctic Amplification (AA, Serreze and Barry 2011). AA is mainly explained by the ICE effect (Perlwitz et al. 2015) displayed in Fig. 9d (ICE induces a warming of approximately 10 °C at the surface, Table 2). The GHG effect dominates the generalized tropospheric warming, particularly over the tropical latitudes, and the stratospheric cooling at high latitudes (Fig. 9c). This is consistent with the IPCC report (Stocker et al. 2014), which suggests that the tropical upper troposphere warming is associated with a decline in the moist adiabatic lapse rate of temperature in the tropics. Though statistically significant, the impact of Arctic sea ice loss is weak in the tropics, with responses weaker than 1 K (Fig. 9d; Table 2). The fact that the ICE effect spreads out in the tropical region is in agreement with DE15, who showed that in presence of ocean–atmosphere coupling, the impact of the Arctic sea ice decline expands to the whole globe. Moreover, we can see in Table 2 that the tropical warming induced by Arctic sea ice loss is very similar for the two coupled models (CNRM-CM5 and CCSM4). The relative GHG and ICE responses induce changes in the meridional and vertical temperature gradients. In particular, the surface meridional temperature gradient decreases due to AA. However, at higher levels, the meridional temperature gradient is enhanced due to stronger warming in the tropics. The consequence of such modifications on the baroclinicity is further investigated in Sect. 3.3. The GHG + ICE UA response (Fig. 9f) reflects a strengthening (more than ~6 m s⁻¹) of the zonal wind speed within the 20–40°N band, and a weaker decrease centered around 60°N. Taking into account the climatological jet core, no latitudinal shift is observed. GHG response dominates the jet stream strengthening above the 500-hPa level. At lower levels, the jet maximum is displaced poleward around 50°N (Fig. 9g). In contrast, ICE effect causes a barotropic weakening of the westerlies on their poleward flank and a slight enhancement in the core of the jet (Fig. 9h). This response is consistent with imprint of the negative phase of the NAM, similarly to DE15. As for the near surface circulation, we show here that the GHG and ICE effects compensate, resulting solely in enhanced westerlies in the core of the jet with no significant latitudinal shifts (DE15; Barnes and Polvani 2015).

Fig. 9

Same as Fig. 7 but for (top) zonal means of the air temperature (TA, units in °C) and (bottom) for zonal means of the zonal wind component (UA, units in m s⁻¹) as a function of pressure levels for winter (DJF) season. The green contours correspond to the climatological temperature (contour interval is 10 °C) and zonal wind (contour interval is 6 m s⁻¹) computed from the CTL20 simulation

In summer (Fig. 10), the GHG + ICE effect for TA displays a generalized warming in the troposphere (stronger in the middle layers over tropical areas) and a slight cooling of the stratosphere (Fig. 10b). These responses are clearly driven by the GHGs increase (Fig. 10c). AA is also noticeable (Fig. 10d) in summer, but much weaker than in winter (~1 versus ~5 K). Once again, Fig. 10d shows that the impact of Arctic sea ice loss expands to the whole Northern Hemisphere, even if ICE effects are small comparing to GHGs effects. The GHG + ICE UA response is characterized by upper zonal wind enhancement, associated with a barotropic northward shift of the zonal wind (Fig. 10f). This is mainly explained by increased GHGs (Fig. 10g). According to the U850 regional changes displayed in Fig. 8f, g, the strengthening and poleward shift of the jet stream in summer seem to be dominated by the North Atlantic. Arctic sea ice loss generates a significant decrease over the core of the jet (Fig. 10h), but this pattern is not visible in the total response. These results are consistent with the northward shift of the jet stream in summer observed in CMIP5 models as reported by Barnes and Polvani (2015).

Fig. 10

Same as for Fig. 9 but for summer (JJA) season

3.4 Wintertime dynamics at synoptic scale

Beyond the position and speed of the jet stream discussed in Sect. 3.2, both GHG and ICE might have an effect on the waviness of the mid-latitude flow, including the occurrence of extreme anticyclonic or cyclonic pressure systems (i.e., blockings or storms) often associated with extreme surface weather. For example Francis and Vavrus (2012) hypothesized that a negative NAM response to AA should be associated with a wavier flow, a hypothesis that has been questioned by several studies (Barnes 2013; Hassanzadeh and Kuang 2015). This section investigates such potential responses to GHG and/or ICE, with a particular focus on the wintertime season when the atmospheric dynamics reaction to Arctic sea ice loss is more intense, and hence the most prevailing for surface weather variability.

3.4.1 Blockings and sinuosity

Here we consider metrics characterizing the large-scale waviness of the mid-latitude flow. A commonly used metric is the frequency of blocking events, i.e., anticyclonic meanders typically associated with a wavy flow. Here we use the blocking index introduced by Tibaldi and Molteni (1990): a particular longitude is said to be “blocked” if the mid-latitude meridional gradient of geopotential height field at 500 hPa (Z500) is locally reversed (see http://www.cpc.ncep.noaa.gov/products/precip/CWlink/blocking/index/index.nh.shtml for more details). Another metric more recently introduced by Martin et al. (2015) is the sinuosity, which intuitively characterizes the flow waviness as the length of the flow trajectory divided by the length of a straight line encompassing such trajectory. Here we use the index proposed by Cattiaux et al. (2016), which consist of the following: For each day, we isolate the flow trajectory from a particular Z500 iso-contour (i.e. isohypse), whose value precisely corresponds to the Z500 average over the latitudinal band 30–70°N, in order to insure that we constantly describe the flow at the same latitude (~50°N). Then we define the sinuosity index as the total perimeter of the isohypse divided by the length of the Earth circle measured at 50°N. In this ways a sinuosity equal to 1 thus corresponds to a perfectly zonal flow at 50°N. Note that this sinuosity metric differs from the blocking frequency: while the latter characterizes synoptic meridional departures from the climatological trajectory, the former includes the waviness of both the mean state and the synoptic activity.

Figure 11 shows the DJF distribution of the number of blocking events averaged over all NH longitudes for CTL20, GHG + ICE, GHG and ICE. We also consider the CTL20 experiment since it represents the climatological values. On average over the 100 years of analysis, the mean frequency is about 6 days per season and per longitude in the CTL20 experiment, with inter-annual variations between 2 and 12 days. The response to both GHG + ICE effects is a significant decrease in the mean frequency of blockings (5 days per season per longitude). At the hemispheric scale, this response is mainly explained by the GHG forcing, since the ICE forcing has no significant effect; however, both GHG and ICE contribute to the decrease in the blocking frequency in the Pacific sector (not shown).

Fig. 11

DJF distributions of (left) the average number of blockings and (right) the sinuosity for all experiments: CTL20, CTL21, ICE20 and ICE21, respectively designated by CTL, GHG + ICE, GHG and ICE to ease the interpretation. Boxplots are drawn from 100 DJF seasons and illustrate the median (thick horizontal segment), the mean (thick dot), the interquartile range (box) and the total range (whisker). Red (blue) asterisks indicate a 10%-level (*) or 5%-level (**) significant positive (negative) difference relative to CTL according to a t test. Green right axes represent departures from the CTL mean (horizontal dotted line) in standard-deviation levels

The mean DJF sinuosity is about 1.4 in the CTL20 experiment (Fig. 11, right panel). We find that the GHG + ICE effects do not significantly alter the sinuosity, due to compensating GHG and ICE individual responses at the hemispheric scale: the GHGs increase tends to significantly reduce the sinuosity, while the sea ice decline is associated with a marginally significant increase in sinuosity (significant at the 90%-level, not 95%). Both responses are mainly due to changes in the North Atlantic sector, where both GHG and ICE effects are significant, while no significant sinuosity change is found in the Pacific (not shown).

Overall, responses of both blockings and sinuosity to GHG and ICE, even when statistically significant, are found to be small relative to the inter-annual variability (less that 0.5 standard deviations in all cases). The most robust signal is found for the GHG response, the strengthening and poleward shift of the Jet Stream being associated with a decrease in waviness. The waviness response for the ICE forcing is less clear at the hemispheric scale, partly due to opposite signals between the Pacific (significant decrease in blocking frequency) and the Atlantic (significant increase in sinuosity).

3.4.2 The Eady growth rate response

As previously shown in Sect. 3.2, the meridional and vertical structures of temperature gradients are modified due to both GHG and ICE effects. These gradients are tightly associated with baroclinic processes leading to cyclogenesis and storm-tracks. We investigate here the respective contributions of GHG and ICE on the synoptic-scale processes during winter. A simple parameter measuring the baroclinic instability is the so-called Eady Growth Rate (EGR) (Lindzen and Farrell 1980; Hoskins and Valdes 1990). This parameter involves the vertical wind shear and the Brunt-Vaisala frequency (measure of the static stability) in the way:

$$\sigma =0.31\times \frac{g}{N\theta }\frac{\partial u}{\partial z}$$

(1)

where N is the Brunt-Vaisala frequency (usually in day⁻¹), θ the potential temperature (in K) and $\frac{\partial u}{\partial z}$ the vertical wind shear (in day⁻¹). Because of the thermal wind balance, the vertical wind shear is connected to the meridional temperature gradient ($\frac{\partial \theta }{\partial y}$). Furthermore, the static stability depends on the vertical gradient of the potential temperature ($\frac{\partial \theta }{\partial z}$).

$$N=~\sqrt{\frac{g}{\theta }\times \frac{\partial \theta }{\partial z}}$$

(2)

It is recommended to compute the EGR at the synoptic scale (i.e., using daily data). It has been computed in CTL20 using daily and monthly timescale and it does not affect the results (not shown). Thus, to reduce computational time, we have decided to show the EGR diagnostic computed from monthly data. It is interesting to quantify the relative contribution of the vertical and meridional temperature gradient to the EGR. Here, we use the same methodology as in Yin (2005) and Graff and Lacasce (2012) to decompose the EGR into two components namely:

$${{\sigma }_{\frac{\partial \theta }{\partial y}}}=0.31~\times ~\left( \frac{g}{{{N}_{CTL}}{{\theta }_{CTL}}} \right)\times ~\frac{\partial \theta }{\partial y}$$

(3)

$${{\sigma }_{N}}=0.31~\times ~\left( \frac{g}{N{{\theta }_{CTL}}} \right)\times ~{{(\frac{\partial \theta }{\partial y})}_{CTL}}$$

(4)

The above decomposition allows estimating the role of the meridional temperature gradient ($\frac{\partial \theta }{\partial y}$) by maintaining the Brunt-Vaisala frequency constant at the climatological value of CTL20. Similarly, to evaluate the role of the vertical temperature gradient ($\frac{\partial \theta }{\partial z}$) embedded in N, the meridional temperature gradient ($\frac{\partial \theta }{\partial y}$) is set to the CTL20 climatology.

The zonal means of the GHG + ICE, GHG and ICE responses in the Northern hemisphere computed for the different contributions (EGR, $\frac{\partial \theta }{\partial y}$, $\frac{\partial \theta }{\partial z}$) are shown in Fig. 12. The total EGR response (Fig. 12a) exhibits an increase of the near-surface baroclinic instability over the Arctic, a decrease at low-level at mid-latitudes (40°N–70°N) linked to the weakening of the mid-latitude jet (Fig. 10f) and an increase in the subtropical mid-to-high troposphere (around 30°N and between 600 and 200 hPa) linked to the strengthening of the subtropical jet (Fig. 10f). The ICE contribution to total EGR is responsible for changes in the near surface and low-level troposphere of the whole Northern Hemisphere (Fig. 12c) linked to the southward shift of the eddy-driven jet (Fig. 10h). On the contrary, the GHG effect dominates the EGR increase in the sub-tropical mid-troposphere.

Fig. 12

Zonal-mean of the winter Eady Growth Rate (EGR, units in day⁻¹) in the Northern Hemisphere. The first column corresponds to the total response (GHG + ICE), the second is for the ICE effect and the third one for the GHG effect. We recall that GHG + ICE effect is computed as CTL21–CTL20, the GHG effect as ICE20–CTL20 and the ICE effect is ICE21–CTL20. For each effect the EGR response is decomposed in (1) changes in EGR due to the meridional temperature gradient $\frac{\partial \theta }{\partial y}$ (second row, units scaled to 10 × day⁻¹); and in (2) EGR changes due to the static stability (last row, units 100 × day⁻¹). Green contours indicate the climatological values computed from the CTL20 simulation (contour interval is 1 day⁻¹). Dotted areas represent statistically significant differences according to a t test at the 5% confidence level

The estimated response of the $\frac{\partial \theta }{\partial y}$ component (Fig. 12d–f) is closely related to the reported changes in the zonal-mean air temperature (Fig. 9). Near the surface, GHG + ICE induces a decrease of $\frac{\partial \theta }{\partial y}$ coherent with AA (Fig. 9a), leading to a weakening of the baroclinic activity below 400 hPa (Fig. 12d). The ICE effect leads to surface decrease of $\frac{\partial \theta }{\partial y}$ that is mainly due to the Arctic sea ice loss (Fig. 12f). However, in the upper levels, EGR increase is mostly controlled by GHGs effect (Fig. 12d, e).

Focusing now on the role of the Brunt-Vaisala parameter (Fig. 12g–i), we observe a near-surface increase of baroclinic activity over the high latitudes, connected to the AA and to a decrease of the static stability and a decrease of the vertical temperature gradient $\frac{\partial \theta }{\partial z}$ (Fig. 10a). Similarly, an increase is noticeable in the tropical upper troposphere that can be explained by the strong warming in the upper troposphere and the stratospheric cooling, that leads to enhance $\frac{\partial \theta }{\partial z}$. The increase of N near the surface (Fig. 12g) is totally explained by the ICE effect (Fig. 12i), while the upper troposphere changes are clearly dominated by GHG.

As a summary, EGR exhibits a near-surface increase over polar latitudes north of 70°N, in response to Arctic sea ice decline, which induces a decrease of vertical potential temperature gradient and hence a reduction of the static stability. EGR low-level decrease between 40–60°N is explained by the weakening of the meridional temperature gradient at low-level, and is entirely due to Arctic sea ice loss. Finally, EGR increase in the subtropical mid-to-high troposphere around 30°N is due to the large upper-level warming induced by the GHG between the tropics and high latitudes.

To go beyond the zonal mean characteristics of the EGR, Fig. 13 shows the regional total EGR response which we vertically integrated from 925 to 400 hPa. In the North Pacific, the GHG + ICE response shows a strengthening and eastward extension of the baroclinicity, and a decrease north of the climatological maximum (Fig. 13a). The GHG effect (Fig. 13c) resembles the GHG + ICE changes but the anomalies are weaker, and shows a latitudinal shift of the jet at the exit region of the latter. ICE effect (Fig. 13e) produces a southward shift of the baroclinicity close to the core region of the jet, consistent with the negative phase of the NAM, as previously shown in Fig. 7d, h. Similar conclusions can be made for the ICE effect in the North Atlantic (Fig. 13f). In the North Atlantic, the GHG + ICE (Fig. 13d) shows a decreased baroclinicity in the climatological maximum region, small increase over the British Isles and a decrease east of Greenland. The GHG effect (Fig. 13d) shows the same pattern with intensified anomalies over the British Isles and Europe and weaker anomalies over the climatological maximum region.

Fig. 13

a, b GHG + ICE, c, d GHG, e, f ICE effects (shading) for the vertically integrated EGR from 925 to 400 hPa in the North Pacific (left) and North Atlantic (right) respectively. We recall that GHG + ICE effect is computed as CTL21–CTL20, the GHG effect as ICE20–CTL20 and the ICE effect is ICE21–CTL20. The green contours correspond to the climatological value computed from the CTL20 simulation (contour interval is 1 day⁻¹). Dotted areas represent statistically significant differences according to a t test at the 5% confidence level

3.4.3 Storm-tracks response

The changes in storm-tracks are expected to be closely related to the meridional temperature gradient (as shown in Harvey et al. 2014, 2015), and hence should be consistent with the EGR parameter.

The Northern Hemisphere storm-tracks are investigated using the tracking scheme developed by Ayrault (1998) based on the detection of maximum of 850 hPa relative vorticity (Flaounas et al. 2016; Sanchez-Gomez and Somot 2016). Once the different trajectories are determined, a density of trajectory map (DT hereinafter) is computed corresponding at each grid point to the density of trajectories in a radius of ~200 km. DT is then weighted by the distance to the central grid point by applying a kernel Gaussian filter.

Figure 14 shows the responses of DT for the North Pacific and North Atlantic, for GHG + ICE, GHG and ICE effects. For clarity, the climatological DT fields estimated from the CTL20 run are also displayed. The climatological maximum of the EGR is located west compared to the maximum storm-tracks for both Atlantic and Pacific, in region of strong local sea surface temperature gradient. In the North Pacific GHG + ICE displays a large-scale structure suggesting a strengthening and eastward extension of the storm-tracks, accompanied by a southward displacement. A slight decrease is observed over lower latitudes (Fig. 14a), which seems to be controlled by the GHG effect (Fig. 14c). On the other hand, ICE dominates the storm-tracks response over the region where the climatological maxima are located (Fig. 14e). The ICE effect is in agreement with the EGR response displayed in Fig. 13e, pointing to a negative phase of the NAM, associated with a southward shift of the North Pacific storm-tracks.

Fig. 14

a, b GHG + ICE, c, d GHG, e, f ICE effects (shading) for density of cyclones tracks (DT in the text) in winter in the North Pacific (left) and North Atlantic (right) respectively. We recall that GHG + ICE effect is computed as CTL21–CTL20, the GHG effect as ICE20–CTL20 and the ICE effect is ICE21–CTL20. The green contours correspond to the climatological value computed from the CTL20 simulation (contour interval is 5 storms per winter). Dotted areas represent statistically significant differences according to a t test at the 5% confidence level

In the North Atlantic, GHG + ICE exhibits a weaker response, with a generalized decrease in the number of extra-tropical cyclones (Fig. 14b). DT decreases over the Great Lakes in North America, the central North Atlantic and the Mediterranean region. A similar pattern has also been found in Zappa et al. (2013), who described a “tripole” structure in the CMIP5 ensemble in response to anthropogenic forcings. This tripole pattern is characterized by a DT decrease over Greenland and Southern North Atlantic, and an increase south of Greenland and west of British Isles. Here, though the tripole structure is also noticeable, only the southern lobe is statistically significant (Fig. 14b). The response due to increased GHGs resembles the GHG + ICE with intensified anomalies in the south of the domain and over the British Isles (Fig. 14d). In contrast, ICE effect produces a weakening of the DT in high latitudes and a slight increase from the Cap Hatteras to the western coast of Europe. The ICE-induced response is again consistent with the negative phase of the NAM (and NAO), as previously shown in Fig. 7d, h from the SLP and U850 anomalies and in Fig. 13f from the EGR response showing an equatorward displacement of the maximum baroclinicity. Even if the DT response is less detectable in the North Atlantic (one reason might be the presence of strong internal atmospheric variability), ICE effect seems to counteract the GHG response, resulting in an almost no significant change in the southern part in GHG + ICE.

4 Summary and discussion

We have investigated the respective roles of direct GHG radiative forcing and induced Arctic sea ice loss on the Northern Hemisphere atmospheric circulation in the CNRM-CM5 coupled model. An idealized experimental protocol has been implemented in order to separate both effects. In such a protocol, based on DE15, a local correction term on the non-solar flux is applied to control the rate of Arctic sea ice formation or melting. The first guess of the value for the flux correction is estimated from the differences between the RCP8.5 (2070–2099) and HIST (1970–1999) periods. Then, a calibration protocol is established in order to obtain, under present GHG concentrations, the Arctic sea ice mean conditions corresponding to the end of the twenty-first century (and vice versa, under future GHG concentrations, the Arctic sea ice conditions for the end of the twentieth century). Comparing to the experimental protocol developed in DE15, there are two important differences: (1) while DE15 used the same flux correction value throughout the Arctic, here the flux correction varies spatially over grid points with Arctic sea ice decrease greater than 10%, (2) we have also performed control simulations with external forcing representative of the last decades of both twentieth and twenty-first centuries to simulate a stabilized climate state for these periods. This allows a comparison between CTL-type and ICE-type experiments (Table 1), and also avoids comparing transitory runs to stabilized sensitivity experiments, which could lead to misleading interpretations.

A validation of the experimental protocol shows that it successfully works to reproduce the future (present) Arctic sea ice conditions under present (future) radiative forcing. The atmospheric adjustment due to flux correction occurs after 40 years of simulations, but the deep ocean circulation (AMOC) stabilizes later (after about 100 years). For this reason, all the experiments have been run for 200 years and only the last 100 years are used in the analysis in this paper. As in DE15, the atmospheric response to Arctic sea ice loss is not confined to the high latitude regions, but expands to the whole globe. The maximum sea ice loss occurs in late summer, but the impact on heat fluxes is maximum in winter (primarily for the latent and sensible heat fluxes). In consequence, the large-scale atmospheric response is also maximum in wintertime. This response, estimated by the ICE isolated effect, is characterized by a southward shift of the Northern Hemisphere westerlies in winter, coherent with the negative phase of the NAM in agreement with other studies (Deser et al. 2010, 2015; Screen et al. 2013). In the North Atlantic, GHG and ICE effects on the mid-latitude westerlies are opposite, leading to a non-significant response in GHG + ICE, by contrast to the North Pacific.

The vertical structure of the Northern Hemisphere zonal mean temperature has also been investigated. ICE effect practically dominates the strong warming over the Arctic Ocean (Arctic Amplification), which vertically extends until the mid-troposphere (500 hPa). By contrast, the GHG impact drives the warming over the subtropical regions, which is maximum around 200–300 hPa, and the cooling at upper levels at high latitudes. These effects are similar in winter and summer, though the responses are enhanced in wintertime. We show that ICE and GHG effects differently affect both meridional and vertical temperature gradients, in particular in winter. These gradients impact the synoptic scale processes in winter, in particular the weather patterns, baroclinicity and the storm-tracks. Our results are in good agreement with the study of McGraw and Barnes (2016) who conclude that an equatorward shift of the jet can be found in response to a polar lower-tropospheric heating, while a poleward shift is induced by a tropical upper-tropospheric heating.

From sinuosity and blocking indices, a significant decrease of sinuosity and blocking events is observed in response to GHGs increase, whereas ICE induces an increase of blocking events, although all changes are small relative to inter-annual variability. As a measure of the baroclinic activity we have used the EGR parameter, which accounts for meridional and vertical temperature gradients. Through a decomposition that allows isolating the vertical and meridional temperature gradients contribution to the EGR zonal mean, we show that the ICE effect is responsible for a decrease in baroclinicity at lower levels in the mid-latitudes, associated with a decrease in the meridional temperature gradient due to Arctic Amplification. On the other hand, there is an increase of EGR at lower levels over the polar latitudes associated with a decrease in static stability (decrease of the vertical potential temperature gradient). GHG also induces EGR changes, which are more confined within the subtropical latitudes (~30°N) at upper levels (from 500 hPa). In particular, an EGR increase, with a maximum at 200 hPa, is reported due to an enhancement of the meridional temperature gradient at upper levels (the warming in the tropics is stronger than in the high latitudes at this level). Near the equator, EGR decreases at 200–300 hPa, due to the weakening of the vertical temperature gradient that increases the static stability. The EGR changes reported here allow quantifying those areas and atmospheric levels where the GHG and ICE effects are dominant. ICE effect on EGR mostly contributes at high and polar latitudes and low atmospheric levels, whereas GHG dominates the impacts at upper levels on subtropical and tropical areas.

Regarding the storm-track response in the North Pacific, ICE effect induces a strong increase of the density of tracks at mid-latitudes, which is coherent with an increase of the EGR parameter vertically integrated between 925 and 400 hPa. GHG effect on the North Pacific storm-track is less relevant and more difficult to interpret. In the North Atlantic, the total response (GHG + ICE) exhibits a tripolar pattern, also documented in Zappa et al. (2013), who studied future projected changes in the North Atlantic storm-track in the CMIP5 models. This tripolar pattern consists of a decrease of storm density in the subtropical latitudes and in the Mediterranean sea, an increase in the British Isles and a decrease over the Greenland and Norwegian seas. With our experimental protocol, we have shown that the increase over the British Isles is mainly explained by the GHGs, while the decrease over the Greenland and Norwegian seas is mainly due to Arctic sea ice loss. Finally the weak decrease in the subtropics can be interpreted by the opposed sign responses between ICE and GHG. This tripolar pattern response is in agreement with the vertically integrated EGR changes. Even if the changes reported here are statistically significant, the storm-track response is more complicated to interpret, probably because of the complex mechanisms involved, and the strong internal variability for that specific field leading to a large uncertainty in the total response (Harvey et al. 2014, 2015) particularly in the North Atlantic region.

This study highlights several prospects. Our results are in good agreement with the study of DE15, which demonstrates that the response to Arctic sea ice loss is fairly robust between CNRM-CM5 and CCSM4 coupled models. This leaves the door open for other models to perform similar experimental protocols. Concerning the flux correction technique applied here, it is important to mention that the methodology is not conservative in terms of energy. We are aware of this weakness and also that the protocol could not allow isolating the effect of Arctic sea ice loss in a physically proper way. In particular, coupled feedback between sea ice and clouds are not represented. An alternative methodology, which allows conserving the energy, consists of modifying the albedo (Blackport and Kushner 2016). In this recent study, the authors investigate the transient and equilibrium response to rapid summer Arctic sea ice loss by decreasing the surface albedo. Such kind of protocol will be tested and explored in a future work. An open and crucial question is the oceanic response to Arctic sea ice loss, in particular the effect on the AMOC and global ocean circulation. The decrease of the AMOC in ICE21 simulation needs undoubtedly further investigation. To investigate properly the effects on the AMOC and on the deep ocean, it should be necessary to perform longer simulation that the ones used in this work. Further analysis, such heat and fresh water analysis on the key regions for the deep convection are part of our perspectives. In this study, we have only investigated the response to Arctic sea ice loss in present climate GHGs concentrations, while we could also evaluate this response with future (end of the twenty-first century) GHGs. This could provide an estimation of the model sensitivity to Arctic sea ice loss under different external forcings. Similarly, we can evaluate the response to GHGs increase with future Arctic sea ice conditions (here we have shown only the response under present Arctic sea ice conditions).

One natural extension of this study is the analysis of surface extreme weather in response to GHG and ICE separated effect. In particular the response of how winter cold extremes over Europe, Asia and North America, are modulated by Arctic amplification is still under debate and no consensus has been found so far. Nevertheless, it has been shown by several studies that sea ice los can reduce their number/severity due to thermodynamical reason (weakened cold air advection) and that this dominates over dynamical considerations (Ayarzagüena and Screen 2016; Screen et al. 2015). Our results are in line with the hypothesis made by Francis and Vavrus (2012) who linked the negative NAM response due to Arctic Amplification to an increase in flow waviness, including an increased frequency of blocking events. But the amplitude of the waviness response to Arctic Amplification in our estimations remains small relative to the inter-annual variability, so that the response of extreme weather, if any, might be hardly discernible.